8th International Seminar on Speech Production 253

Anticipatory Coarticulation in a Specific Context

Gósy, Mária – Beke, András Institute for Linguistics, Hungarian Academy of Sciences – ELTE University, Budapest E-mail: [email protected] – [email protected]

Abstract compensatory lengthening in the preceding takes place before non-continuant stops and In Hungarian casual speech, [n] may not be fully . The spreading coarticulatory effect in the pronounced before continuant . In this VNC sequences is supposed to originate from the paper the durational changes of the vowel and the continuant C that modifies the [n] articulation. Our nasal as well as the formants of the vowel were hypothesis is that the modified nasal affects not analyzed in VNC sequences. The spreading only the duration but also the formant frequencies coarticulatory effect in these sequences is supposed of the preceding . We also suppose that both to originate from the continuant C ( and the fricatives and the will affect the approximants) that modifies the [n] articulation. preceding sounds in the analyzed VNC sequences. The results show that there is a strong spreading The goal of the present research was to show the anticipatory coarticulation leading to significant acoustic-phonetic consequences of spreading antici- changes in the duration of the vowels and the nasals patory coarticulation in terms of nasal duration and preceding fricatives and approximants. The coar- the duration and formant frequencies of the preced- ticulation affects the F2-value of the vowel as well. ing vowels.

1 Introduction 2 Subjects, material, method

Speech is produced as a sequence of sounds as Ten native female speakers of Hungarian with no the vocal tract moves from one articulatory known speech or hearing defects read randomized configuration to the next. Coarticulation refers to isolated words and phrases in a sound-proofed the transition from one gesture to another whereby chamber (ages ranged from 20 to 28). the different articulation gestures combine with The word list consisted of 20 Hungarian words different timing patterns [2, 3]. Research has shown and phrases that contain the dento-alveolar nasal [n] the existence of various effects in followed by either a , or a stop/ or VowelNasalConsonant sequences across languages [1, an . The VN cluster belonged to the 5], while less attention has been paid to first syllable in all of the words, while the second compensatory lengthening in the vowels in such syllable started with the other as in the sequences depending on the type (continuant vs. word kAN/Dúr (‘tomcat’). The [] and [ ] vowels non-continuant) of the consonant following the were chosen to precede the nasal sound because nasal sound [4]. In this study spreading anticipatory they have no phonologically long counterparts, on coarticulation in Hungarian was analyzed in the the one hand, and they are the most frequent vowels sequences in Hungarian, on the other hand. Examples: kancsal (‘cross-eyed’), hentes (‘butcher’), lenzsák (‘flax bag’), tanszék (‘department’), kancát (‘mare’ +Acc.), lencsét (‘lentil’+Acc.), fenség (‘majesty’). The vowels are followed by the dento-alveolar nasal The words were recorded in a sound tight sound and the post-nasal C which is either an chamber and digitalized up to 44,000 Hz. Acoustic- (stop, fricative or affricate) or an phonetic analysis was carried out by Praat software, approximant. In Hungarian casual speech, /n/ may version 5.4. The nasal [n], the previous vowel, and not be fully pronounced before continuant fricatives the following consonant were identified in each and approximants [6]. Closure for the nasal may be word for each speaker. The duration of the nasal incomplete, resulting in a shorter nasal sound and in consonant and of the preceding vowel, as well as of this case the preceding vowel is lengthened. Neither the first three formants of the vowels were the modified articulation of the nasal, nor measured. The duration of the vowels was measured

ISSP 2008 8th International Seminar on Speech Production 254

as the interval from the second formant onset to the onset of the nasal formants. The corresponding spectrographic and waveform displays were consulted and auditory perception was also considered. Nasal duration was measured to the onset of the obstruent consonant and to the onset of the second formant of the approximant (Fig. 1.) The formant structure of the vowels was measured by means of the series of frequency values measured in every 0.006 ms of the total duration of the vowel. The formant values were normalized. This method provided an opportunity to analyze the tendency of [k n t l] changes in frequency patterns for the whole duration of the vowel. Statistical analysis of the data was carried out using ANOVA and regression analysis (using SPSS 14.0). In all cases, the confidence level was set at the conventional 95%.

3 Results

Acoustic-phonetic analysis confirmed our expectation that spreading anticipatory coarticula- tion appears in VNC sequences. Continuant and approximants affect the duration of nasals, making them shorter compared to nasals preceding non-continuant obstruents. The tempo- [t n s e k] rally modified (shortened) nasal has a twofold effect on the preceding vowel: (i) the vowel is lengthened, and (ii) its second formant frequencies will be different from those preceding nasals followed by non-continuants. 3.1 Coarticulation effects on the nasal The acoustic structures of the nasal [n] preceding continuants and non-continuant s can be clearly seen in the examples shown in Figure 1. The nasal sounds are relatively longer before stops and affricates than before fricatives and approximants (Fig. 2). The mean duration of [n] before non- continuants is 104.95 ms (SD=23.75), 59.55 ms

(SD=25.74) before continuants, and 81.91 ms [k n l o ] (SD=18.36) before approximants. The statistical analysis showed significant differences in nasal Figure 1: The acoustic structure of the words durations depending on the type of the following kancsal (‘cross-eyed’), tanszék (‘department’) and consonant (one-way ANOVA: F(2, 199) = 114.860, kanló (‘stallion’) p<0.001). In some languages, nasals are shorter before voiceless than before voiced consonants [1]. The mean value of the duration of [n] is 128.35 This fact also can be seen in our data; however, it ms (SD=18.28) when followed by voiced stops and could be statistically confirmed only for stops in 102.35 ms (SD=16.21) when followed by a Hungarian. As expected, the duration of [n] is voiceless stop. The duration of the nasal also shows significantly shorter depending on voicing in the a tendency to become shorter before voiceless following stops (one-way ANOVA: F(1, 79) = fricatives. The mean value of [n] is 63.57 ms 37.994, p<0.001); however, there is no such (SD=23.99) before a voiced fricative and 56.97 ms difference in the case of the following fricatives. (SD=25.92) before a voiceless fricative (one-way

ISSP 2008 8th International Seminar on Speech Production 255

ANOVA: F(1, 79) = 1.621, p<0.207). In sum, mean = 146.3 ms, SD=25.46; for []: mean = 161.05 continuant obstruents affect the temporal patterns of ms, SD=22.62). The shortest duration was found nasal articulation while the approximants have less when the nasal was followed by non-continuants pronounced, but similar effects on them. (for [ ]: mean = 132.1 ms, SD=15.20; for []: mean = 135.4 ms, SD=28.00), cf. Fig. 3. Statistical 175 analysis proved that there was a significant difference in durations depending on the three 150 consonant types (one-way ANOVA for [ ]: F(2, 79)

125  20 = 10.855, p<0.001 and for [ ]: F(2, 79) = 33.734, 1727

) p<0.001).

ms 100 (

220 75 3 duration

200 50

180 25

60 ) 160

0 ms ( non-continuants continuants [l] and [j]

140

Figure 2: Durational data of the nasal [n] as a duration function of the following consonant 120 (median and range)

100 The shows some stability before 8 non-continuants (stops and affricates) which means 80 non-cont cont appr non-cont cont appr that the nasal preserves the closure in its articulation. Therefore the duration of the nasal is [ ] vowels [] vowels longer. The [n] is less stable before fricatives. In Figure 3: Durational data for the vowels [ ] and [] these cases the articulation of the nasal is in various VNC contexts (median and range) characterized by an early onset of the velum gesture and by the lack of closure. The blade of the tongue It was hypothesized that in addition to durational approaches the alveolar ridge but never reaches it. changes, the formant structure of the vowels would This “irregular” articulation gesture that does not also be affected depending on the type of the obstruct the airflow in the vocal tract results in a consonant following the nasal in the sequence. The shortened nasal. This may be regarded as an data in Table 1 show that there are no great anticipatory coarticulation effect of continuants. differences in the first and third formant values of Since fricatives and the [l] and [j] sounds share the the vowels depending on continuancy or non- same „friction” gesture in their articulation, continuancy of the obstruents or approximants. The approximants may be expected to show a similar F2 values of the vowels, however, show significant coarticulatory effect. Indeed, the durational data differences, depending on the consonant that confirmed that, although approximants seem to follows the nasal (one-way ANOVA for [ ]: F(2, form a separate category in respect of coarticulatory 2165) = 66.728, p<0.001 and for []: F(2, 2022) = effect, the duration of the nasal before approximants is very similar to that before continuant obstruents. 39.116, p<0.001). The post-hoc Tukey test confirm- ed significant differences for all the consonants 3.2 Coarticulation effects on vowels studied. Continuant obstruents and approximants The duration of the vowels preceding the nasal have the same, while non-continuant obstruents sound was expected to be longer before continuants have different effects on the vowels. than before non-continuants. The duration of both The mathematical functions applied also support the analyzed vowels was the longest when the nasal the finding that the type of C in VNC sequences is followed by a continuant obstruent (for [ ]: mean results in different coarticulatory effects. Figure 4 shows the values of the second formants in the time = 158.5 ms, SD=29.74; for []: mean = 178.7 ms, course of the [ ] vowels in the words tanszék SD=18.91). A shorter duration was found when the (‘department’) and kancsal (‘cross-eyed’). nasal was followed by the approximants (for [ ]:

ISSP 2008 8th International Seminar on Speech Production 256

Table 1. Mean values of F1, F2 and F3 for the 236 ms. The durations of VN sequences followed vowels preceding the nasal. by approximants show similar values at 221 ms vs. 243 ms (Table 2). Thus, vowel lengthening as a F C [ ] [] spreading coarticulatory effect in the VNC mean mean sequence can be viewed as a coarticulatory shift SD SD (Hz) (Hz) within the VN combination. non- 712.8 149.9 670.1 115.3 cont. F1 Table 2. Total duration of VN combinations. cont. 673.9 183.7 645.7 133.0 appr. 756.2 143.4 632.0 116.1 Data (ms) [ ] + [n] [] + [n] non- 1322.2 207.7 1891.7 277.6 obstruent [l, j] obstruent [l, j] cont. F2 Mean 236.68 221.7 236.2 243.3 cont 1387.4 223.4 1839.7 247.6 SD 18.19 25.2 19.9 33.9 appr. 1474.4 202.1 1984.0 302.4 non- 2630.9 357.1 2758.8 417.8 Spreading anticipatory coarticulation works cont. F3 across syllable boundaries in Hungarian. It first cont. 2640.6 429.0 2762.3 426.8 affects the nasal’s duration, and then it exerts an appr. 2664.7 419.4 2814.4 305.0 effect on the duration and the second formant of the Abbreviations: F – formants, C – consonants following the nasal, non-cont. – non-continuants, cont. – continuants, appr. – preceding vowel. The tongue body’s raising or approximants lowering gesture seems to have an impact on F2. The effect of the voiced stop consonants is A linear function was applied to the F2-values in manifested in a significantly longer duration of the cases where C was either a stop or an affricate, and nasal compared to that of voiceless stops. However, a polynomial function was used where C was either this is not the case with fricative consonants. The a fricative or an approximant (an example for the coarticulatory effect was also confirmed for the latter is kanló /’stallion’/: y=0.9052x2 – 19.997x + approximants: this appeared to be more similar to 1038.8, R2 = 0.9551). that caused by continuant obstruents than by non- continuants. This research showed the existence of 1800 a strong spreading anticipatory coarticulation y/tanszék/ = 2.6876x2 - 63.06x + 1747.1 1700 R2 = 0.9707 process resulting in the modification of the 1600 articulation of vowels and nasals preceding

1500 fricatives and approximants. The issue of antici-

z patory behavior is crucial for the understanding of

H 1400 speech production. The present results reflect the 1300 domain over which the motor system organizes 1200 VNC sequences of articulation. y/kancsal/ = 12.982x + 1207.1 kancsal 1100 2 = 0.8096 tanszék 1000 References 6 5 0 6 5 12 18 2 31 37 43 5 5 62 68 7 81 87 93 100 106 ms [1] P. S. Beddor. Nasals and nasalization: The relation Figure 4: Functions applied to the F2-values for between segmental and coarticulatory timing. In Proc. of two Hungarian words 16th ICPhS, Saarbrücken, 249–255, 2007. [2] S. E. Boyce, R. A. Krakow, F. Bell-Berti, C. E. 4 Conclusions Gelfer. Converging sources of evidence for dissecting ar- ticulatory movements into gestures. Journal of Analysis of the durational patterns of nasals and 18: 173–188, 1990. [3] M. Chafcouloff, A. Marchal. Velopharyngeal vowels in VNC sequences raises the question coarticulation. In W. Hardcastle, N. Hewlett (eds.), whether what has so far been regarded as Coarticulation. Cambridge University Press, Cambridge, compensatory lengthening is rather a compensatory 69–79, 1999. shift within the vowel+nasal combination. In this [4] D. Kavitskaya. Compensatory lengthening: pho- case the total duration of the combination should be netics, phonology, diachrony. Routledge, London, 2002. reasonably stable. Our data show that the total dura- [5] P. Ladefoged. A course in phonetics. Harcourt, Brace, tions of [ n] and [n] sequences, independently of and Jovanovich, Fort Worth, Texas, 1993. [6] P. Siptár, M. Törkenczy. The phonology of Hun- vowel quality and type of obstruent following the garian. Oxford University Press, Oxford, 2000. nasal, show very similar values, amounting to about

ISSP 2008